June 2, 2026
Philip J. Armstrong walked off the BrightonSEO main stage straight into our podcast booth. Friday, May 1, 2026. Three days earlier, on April 28, Adobe had completed its $1.9 billion acquisition of Semrush.
So when Diantha van Surksum and I sat down with him to record, Philip was, by definition, a brand-new Adobe employee with strong opinions about an attribution model the rest of the industry is still trying to defend.
He had just delivered a talk called “Search is the largest behavioral dataset in marketing.” The slides ran on the data behind what he and the BrightonSEO talk page both call the searchpocalypse. The opening claim was the one that should be sitting in every SaaS marketing lead’s deck next quarter:
80% of consumer journeys contain influential AI waypoints. 71% of AI users say it actively shapes their final purchase decision.
Philip J. Armstrong, BrightonSEO April 2026 talk abstract
And here is the gap that makes that number a budget fight rather than a fun stat: when those same brands open their own log files, they see roughly 1% of inbound traffic referred from AI sources.
One percent in the logs. Eighty percent in the journeys. Both numbers are real. The space between them is where the GEO budget conversation is being lost.
Why does AI look like a 1% channel in GA4?
Because that is the part that the referrer header can see. When a user clicks a link in ChatGPT or follows a citation in an AI Overview, the resulting hit shows up with a recognizable referrer and lands in your "AI sources" bucket.
When that same user reads the AI's answer, doesn't click, then opens a new tab and types your brand into the address bar three days later, the visit lands in "direct." When they bounce through Reddit, then Perplexity, then back to Google, then to Amazon before buying, your dashboard sees the last touch and credits it.
Philip's framing was sharper than that:
The 1% is from his client conversations. The 80% (which he rounded up to "almost ninety" in person, but lives on the published slide as 80%) is from Semrush's behavioral panel. I've heard the 1% from several agency owners. They show me the dashboard, they shrug, and they ask why anyone is investing in GEO. They are not lying. They are looking at one piece of evidence and concluding the whole story.
The story is in the gap.
Did direct-click attribution ever actually work?
I pushed back on Philip during the recording. The instinct of probably most SEO leads was that pre-AI attribution was a solved problem. Click came in, conversion landed, and the credit was easy to assign. Hard now, easy then.
He disagreed, and the disagreement matters:
The argument is that the model worked when the customer journey had two real steps. A click, a buy. The moment a journey involved any prior consideration (review sites, comparison pages, a friend's recommendation, a podcast, a YouTube explainer), the last click stopped being the decision and became the receipt.
We accepted it because Google handed us a metric, the metric was free, and the alternative meant doing actual measurement work.
Then AI compressed the consideration step from a multi-day browse into a 90-second answer. The receipt stopped showing the journey. We noticed.
Philip's twenty-year-old solution is unchanged, and it is the part most SEO leads keep refusing to do:
Turn off the channel for a defined window. Hold out a control group. Measure the delta. Marketing mix modeling at a small scale. He spent his Nike and Sony years running exactly that test, and his caution is worth repeating: don't turn it off for too long, especially on brand. The hole gets harder to climb out of the deeper it goes.
The harder version of "we can't attribute AI" is "we never could attribute search, we just got comfortable with a proxy that AI happened to break first."
Ready to start scaling your business?
Let's get started
Why is observed clickstream data different from what most tools show?
The 80% number is doing a lot of work. So it matters where it comes from.
Semrush's published Traffic & Market panel runs on clickstream data from a network of over 200 million real, anonymized internet users across more than 190 countries, with billions of events flowing in monthly.
Philip rounded down to "twenty million active users, twenty billion events" in the conversation. The numbers I can verify from Semrush's own knowledge base are bigger. Either way, the unit of measurement matters less than the type of measurement.
Most attribution data you see in a SaaS dashboard is some flavor of modeled. Modeled-attribution data starts with a small panel, then multiplies it to a national projection. Modeled conversion data fills in the gaps where consent or tracking blocked the raw event. Philip called the underlying class of this work synthetic data, and gave us the single sharpest line of the whole conversation:
Synthetic data sounds scientific. That is the problem. The word lends authority that the data has not earned. There are use cases for it. There are also a lot of decks running on synthetic data dressed up as observation.
The Semrush journey data Philip ran his talk on is the second kind. Behavioural observed. A real anonymised user, a real sequence of sessions, a real conversion event. No model in the middle. That distinction is what lets him stand on a stage and say "AI is in 80% of conversion journeys" rather than "our model estimates AI is probably in some journeys at some rate." The reader's job is to ask the same question of every attribution claim landing in their inbox this quarter: is this observed, or is this modelled?
When the answer is "modelled," ask what it was modelled from. When the answer is "observed," ask how large the panel is and whether the sources are diverse. Both questions get you further than the dashboard does.
What does it mean when AI is a customer of your data?
The third axis Philip pulled in is the one most SEO teams have not started planning for. Diantha asked him about Semrush's vision, and he gave us a sentence that should be the first slide of any GEO budget defence in 2027:
Read that twice. It reframes the entire data stack. The buyers of your insights, dashboards, and reports are not just humans clicking through tabs.
The buyers include the agents who will increasingly run between your tools, your sales process, and your customers' decision-making. If your data isn't accessible to an agent (whether through MCP, an API, or whatever standard wins), it won't be accessible for the next decade of marketing operations.
I have seen this from the other side. Every SEO agency owner I talk to is wrestling with the same question: do we let our clients' agentic stacks read directly from our tooling, or do we keep playing intermediary? The first answer is uncomfortable, and the second answer is a slower way to get there. Philip's framing collapses the choice. The data infrastructure that wins the next ten years is the one that treats agents the same way the last ten years treated dashboards.
That is also what changes the meaning of the 80% number. AI, as a waypoint, is on the consumer side. AI, as a customer of data, is on the operator side. The same trend appears on both sides at once.
Ready to start scaling your business?
Let's get started
What happens when the data is there but isn't trusted?
We asked Philip for his biggest marketing mistake. He gave us Nike running. He was clear he wasn't the one responsible, but he saw it:
Independent third-party data backs the anecdote. Between March 2022 and March 2024, Similarweb's clickstream data showed Nike.com's running-category traffic declined steadily, while Hoka grew 59% year over year, Brooks 28%, and Asics 25%. By 2024, Brooks's running site was rivaling Nike's running category for raw traffic. The market noticed in 2024. The search data, by Philip, was known earlier.
The mistake wasn't analytical. It was political. Big companies have brand surveys; the brand surveys disagreed with the search data, yet they were the trusted instrument. Search was a "supporting" signal, easy to discount when it told an uncomfortable story.
That is the deeper version of the AI attribution gap. The 1% in your logs and the 80% in your journeys aren't really competing facts. They are competing instruments. One is convenient and historically trusted, while the other is harder to read and not yet fully credentialed within your organization.
The pattern that costs Nike running will lose your GEO budget next quarter if you let it. The instrument that tells the uncomfortable story gets dismissed, and by the time the comfortable instrument finally tells the same story, the market has moved.
What changes this week?
Two practical things came out of the conversation.
First, stop defending GEO budget with the referrer number alone. Defend it with a journey measurement. Pick one observed signal you are willing to anchor your reporting to, then use it consistently.
Branded search lift, organic-direct ratio, and post-AI-citation cohort behavior. Anchor the budget to it. The referrer number is supporting evidence, not the whole story.
Second, when you cannot get observed data, run the test Philip would. Turn it off. Pick one segment, one geography, one product line, kill the AI-touchpoint investment for four weeks, and watch what happens to the conversion curve.
Marketing mix modeling for people without a marketing mix modeling budget. The principle has been the same since before Google handed anyone a free metric: if you want to know the incremental impact, remove the input.
The Reframe nobody wants to print yet: AI didn't break attribution. The attribution model we relied on was always a proxy. We loved the proxy because it was free. AI just made the gap visible enough for the proxy to stop paying our salaries.
The teams that survive the next twelve months are the ones who admit that and rebuild the measurement layer, not the ones that wait for Google to hand them another easy metric.
The data is there. Philip's job is showing it on stages. Yours is escalating it.
Resources
- The full SEO Cast episode with Philip J. Armstrong. Spotify
- Philip J. Armstrong's BrightonSEO April 2026 session, "Search is the largest behavioral dataset in marketing." brightonseo.com
- Adobe has completed its acquisition of Semrush on April 28, 2026. Adobe Newsroom
- Semrush Traffic & Market panel methodology. Semrush Knowledge Base
- Similarweb on Nike's running-category traffic decline, May 2024. Yahoo Finance / Sourcing Journal
- Philip J. Armstrong, BrightonSEO speaker profile. brightonseo.com/people
Listen to the full conversation
The full episode is live on SEO Cast. "BrightonSEO: Philip J. Armstrong on the death of click attribution and the real consumer journey".
Head of SEO
As Head of SEO at Seeders, Hans helps companies scale smarter by combining search behaviour, paid performance, and conversion insights into systems that drive real growth. Recognised as a Top 20 Content Marketer by Small Business Trends, he brings a lean, experimental approach focused on building what actually works. Rather than relying on reports or one-off strategies, Hans focuses on creating marketing systems and tools that turn visibility into measurable business impact. His goal is simple: help clients move from scattered efforts to structured growth.
